#!/usr/bin/env node
/**
 * checkTerms.mjs — PLAN.md §16, D63 and D64. The twelfth check, added in phase 14.
 *
 * D63: outside a game's own section, the site may NAME the games it supports, but it may not
 * use their VOCABULARY. Core concepts keep core's names — a Team, a game server, the sidecar,
 * the bridge plugin, the game module — because that is what core itself calls them, and a
 * site that says "shard" on a core page is describing one game's deployment as though it were
 * the platform.
 *
 * D64: a convention that holds only while everyone remembers it does not hold. This is the
 * same argument `checkTokens.mjs` makes for colours, applied to words. The vocabulary problem
 * was measured before this check existed — "shard" 227 times in 45 files — and none of that
 * was anybody's decision; it was the residue of a site written while there was one game.
 *
 * What it reads, as of phase 14:
 *
 *   - every page under `src/content/docs`, frontmatter included (a title and a description
 *     are what a reader sees first, in the sidebar and in search results);
 *   - `src/config/sidebar.mjs`, as DATA rather than as text: every group label and every
 *     page label, in both trees. Its comments are not scanned, because a comment is not
 *     something a reader of the site ever sees.
 *
 * Phase 17 widens it to all of `src/`, once the marketing pages have a game page to move
 * their game content to. That is also when it learns D67's screenshot tag.
 *
 * Where a game's words are allowed:
 *
 *   - in that game's own docs directory (`GAME_DIRS`), and its own sidebar group — a group
 *     whose label is the game's `name` in `platform.json`;
 *   - between `<GameExample game="…">` tags on a core page (D68). The component prints the
 *     game's name above the block, so the words arrive already labelled.
 *
 * What is NOT read, deliberately:
 *
 *   - link targets and bare URLs. A repository is called what it is called, and an address is
 *     not prose. The LINK TEXT is read, so `[Module-uo](…)` fails and `[releases](…)` passes;
 *   - `import` lines and `{…}` expressions. They are code, and what an expression renders is
 *     a value from `platform.json` or `src/data/`, which phase 17 brings into scope.
 *
 *   node scripts/checkTerms.mjs
 *
 * No token and no network: everything it reads is in this repository.
 */

import { readFileSync } from 'node:fs';
import { readdir } from 'node:fs/promises';
import { fileURLToPath } from 'node:url';
import path from 'node:path';

import { docsSidebar, plannedSidebar, flatLabels } from '../src/config/sidebar.mjs';

const ROOT = fileURLToPath(new URL('..', import.meta.url));
const platform = JSON.parse(readFileSync(path.join(ROOT, 'src/data/platform.json'), 'utf8'));

/**
 * The words, each tagged with the game it belongs to. Matching is case-insensitive and on
 * word boundaries, so "shard" does not match `SHARD_VISIBILITY` and "wipe" does not match
 * "wiped" unless it is listed.
 *
 * A word is on this list because it is a game's name for something core has its own name for
 * (shard → game server, guild and clan → Team), or because it names one game's component
 * (uo-link → the sidecar, ServUO → the game server). Game NAMES — Ultima Online, UO, Rust —
 * are not on it: D63 allows them everywhere. Steam is not on it either: Steam sign-in is a
 * core identity provider, even though Rust was its first user.
 */
const TERMS = [
  { game: 'uo', word: 'shard', re: /\bshards?\b/gi },
  { game: 'uo', word: 'ServUO', re: /\bservuo\b/gi },
  { game: 'uo', word: 'uo-link', re: /\buo-?link\b/gi },
  { game: 'uo', word: 'Module-uo', re: /\bmodule-uo\b/gi },
  { game: 'uo', word: 'Bridge.cfg', re: /\bbridge\.cfg\b/gi },
  { game: 'uo', word: 'cliloc', re: /\bclilocs?\b/gi },
  { game: 'uo', word: 'guild', re: /\bguilds?\b/gi },
  { game: 'uo', word: 'IDOC', re: /\bidocs?\b/gi },
  { game: 'rust', word: 'rust-link', re: /\brust-?link\b/gi },
  { game: 'rust', word: 'Rust-Plugins', re: /\brust-plugins\b/gi },
  { game: 'rust', word: 'Module-Rust', re: /\bmodule-rust\b/gi },
  { game: 'rust', word: 'Oxide', re: /\boxide\b/gi },
  { game: 'rust', word: 'Carbon', re: /\bcarbon\b/gi },
  { game: 'rust', word: 'uMod', re: /\bumod\b/gi },
  { game: 'rust', word: 'clan', re: /\bclans?\b/gi },
  { game: 'rust', word: 'wipe', re: /\bwipes?\b/gi },
];

/** Each game's own docs directory. A page under one may use that game's words. */
const GAME_DIRS = {
  uo: 'src/content/docs/docs/ultima-online/',
  rust: 'src/content/docs/docs/rust/',
};

/**
 * The allowlist: an exact phrase in one file that may use a word the rule would otherwise
 * refuse. Short on purpose, and every entry says why. An entry that no longer matches its
 * file fails the check too, so the list cannot rot into a set of permanent exemptions.
 */
const ALLOW = [
  {
    file: 'src/content/docs/docs/administration/notifications-and-email.mdx',
    text: 'Guild (server) ID',
    why: "Discord's own word, not a game's: it is the label on core's Discord Bot screen.",
  },
];

for (const game of Object.keys(GAME_DIRS)) {
  if (!platform.games[game]) throw new Error(`checkTerms: GAME_DIRS names "${game}", which platform.json does not.`);
}
for (const t of TERMS) {
  if (!platform.games[t.game]) throw new Error(`checkTerms: "${t.word}" is tagged "${t.game}", which platform.json does not know.`);
}

const failures = [];
const fail = (where, detail) => failures.push({ where, detail });

/** Replace a span with spaces, keeping newlines, so later line numbers stay true. */
const blank = (text) => text.replace(/[^\n]/g, ' ');

/** The game whose directory holds this file, or null for a core page. */
function gameOfPath(relative) {
  const posix = relative.split(path.sep).join('/');
  for (const [game, dir] of Object.entries(GAME_DIRS)) if (posix.startsWith(dir)) return game;
  return null;
}

/**
 * Turn an MDX source into the text a reader would actually see, with everything that is not
 * prose blanked out. Returns that text plus the GameExample regions, as offsets into it.
 */
function proseOf(source, relative) {
  let text = source;

  // GameExample regions first, while the tags are still intact. Recorded, then the tags
  // themselves are blanked so the attribute does not read as prose.
  const regions = [];
  for (const m of text.matchAll(/<GameExample\s+game="([^"]*)"\s*>([\s\S]*?)<\/GameExample>/g)) {
    const game = m[1];
    if (!platform.games[game]) {
      fail(`${relative}:${lineOf(source, m.index)}`, `<GameExample game="${game}"> names a game platform.json does not have`);
    }
    regions.push({ game, start: m.index, end: m.index + m[0].length });
  }
  text = text.replace(/<\/?GameExample\b[^>]*>/g, blank);

  text = text.replace(/<!--[\s\S]*?-->/g, blank); // HTML comments
  text = text.replace(/^import\s.*$/gm, blank); // MDX imports
  text = text.replace(/\]\([^)\s]*\)/g, (m) => '] ' + blank(m.slice(2))); // link targets, not their text
  text = text.replace(/\b(?:href|src)=(?:"[^"]*"|\{`[^`]*`\})/g, blank); // JSX/HTML link targets
  text = text.replace(/\bhttps?:\/\/[^\s)>"'`]+/g, blank); // bare URLs

  // `{…}` expressions, outside code fences only — a fenced JSON example's braces are prose.
  // Innermost first, repeatedly, so nested and multi-line expressions both go.
  const parts = text.split(/(^```[\s\S]*?^```)/m);
  for (let i = 0; i < parts.length; i += 2) {
    let prev;
    do {
      prev = parts[i];
      parts[i] = parts[i].replace(/\{[^{}]*\}/g, blank);
    } while (parts[i] !== prev);
  }
  text = parts.join('');

  return { text, regions };
}

function lineOf(source, index) {
  return source.slice(0, index).split('\n').length;
}

/** Hits for every term the given context may not use. */
function scan(text, allowedGame) {
  const hits = [];
  for (const t of TERMS) {
    if (t.game === allowedGame) continue;
    t.re.lastIndex = 0;
    for (const m of text.matchAll(t.re)) hits.push({ term: t, value: m[0], index: m.index });
  }
  return hits;
}

async function* walk(dir) {
  let entries;
  try {
    entries = await readdir(dir, { withFileTypes: true });
  } catch {
    return;
  }
  for (const entry of entries) {
    const full = path.join(dir, entry.name);
    if (entry.isDirectory()) yield* walk(full);
    else if (/\.mdx?$/.test(entry.name)) yield full;
  }
}

// ── 1. The documentation pages ──────────────────────────────────────────────
let files = 0;
const allowUsed = new Set();

for await (const file of walk(path.join(ROOT, 'src', 'content', 'docs'))) {
  const relative = path.relative(ROOT, file).split(path.sep).join('/');
  const posix = relative;
  const source = readFileSync(file, 'utf8').replace(/\r\n/g, '\n');
  files++;

  let { text, regions } = proseOf(source, relative);

  for (const [i, entry] of ALLOW.entries()) {
    if (entry.file !== posix) continue;
    if (text.includes(entry.text)) {
      allowUsed.add(i);
      text = text.split(entry.text).join(blank(entry.text));
    }
  }

  const pageGame = gameOfPath(relative);
  for (const hit of scan(text, pageGame)) {
    const region = regions.find((r) => hit.index >= r.start && hit.index < r.end);
    if (region && region.game === hit.term.game) continue;
    const where = region ? `inside <GameExample game="${region.game}">` : pageGame ? `in the ${platform.games[pageGame].name} group` : 'on a core page';
    fail(
      `${relative}:${lineOf(source, hit.index)}`,
      `"${hit.value}" is ${platform.games[hit.term.game].name} vocabulary, ${where}`,
    );
  }
}

for (const [i, entry] of ALLOW.entries()) {
  if (!allowUsed.has(i)) {
    fail(`allowlist: ${entry.file}`, `"${entry.text}" no longer appears there — remove the entry`);
  }
}

// ── 2. The sidebar, as data ─────────────────────────────────────────────────
const gameByName = new Map(Object.entries(platform.games).map(([key, g]) => [g.name, key]));
const trees = [
  ['docsSidebar', docsSidebar.map((g) => [g.label, flatLabels(g.items)])],
  ['plannedSidebar', Object.entries(plannedSidebar).map(([label, items]) => [label, flatLabels(items)])],
];
let labels = 0;
for (const [treeName, groups] of trees) {
  for (const [groupLabel, items] of groups) {
    const game = gameByName.get(groupLabel) ?? null;
    for (const label of [groupLabel, ...items]) {
      labels++;
      for (const hit of scan(label, game)) {
        fail(
          `src/config/sidebar.mjs ${treeName} → ${groupLabel}`,
          `"${label}" uses "${hit.value}", ${platform.games[hit.term.game].name} vocabulary, ${game ? `in the ${platform.games[game].name} group` : 'in a core group'}`,
        );
      }
    }
  }
}

// ── Report ──────────────────────────────────────────────────────────────────
if (failures.length === 0) {
  console.log(
    `checkTerms: ${files} docs pages and ${labels} sidebar labels use only their own game's words ` +
      `(${TERMS.length} terms, ${ALLOW.length} allowlisted phrase${ALLOW.length === 1 ? '' : 's'}).`,
  );
} else {
  console.error(`\ncheckTerms: ${failures.length} game word(s) outside their game's section:\n`);
  for (const f of failures) console.error(`  ✗ ${f.where}\n      ${f.detail}`);
  console.error(`
PLAN.md §16, D63: name the games anywhere, but keep their vocabulary in their own section.
Use core's word instead — a Team, a game server, the sidecar, the bridge plugin, the game
module. If a core page genuinely needs a game's real example, wrap it in
<GameExample game="…"> (D68), which labels it with the game's name. Add to the allowlist
in this script only for a word that is not a game's at all, and say why.
`);
  process.exit(1);
}